Please wait a minute...
Data Analysis and Knowledge Discovery
Current Issue | Archive | Adv Search |
Question Classification Based on Hierarchical Attention Multi-Channel Convolution Bidirectional GRU
YU Bengong,ZHU Mengdi
(School of Management, Hefei University of Technology)
(Key Laboratory of Process Optimization & Intelligent Decision-making, Ministry of Education, Hefei University of Technology, Hefei 230009, China)
Download: PDF(995 KB)  
Export: BibTeX | EndNote (RIS)      
Abstract  

[Objective] Question classification is the basic task of question answering system. Its accuracy directly affects the overall performance of the system. In view of the short length and sparse features of dialogue questions, multi-level features extraction of the question text can understand the semantics of questions more comprehensively and effectively improve the classification accuracy.

[Methods] In order to enrich the semantic representation of the problem text and fully consider the interrogative words, part of speech and word position features in the question sentences, multi-channel attention feature matrices are constructed based on the multi-feature attention mechanism at the word level. Secondly, the multi-channel feature matrices are convolved to obtain phrase-level feature representation. Then the vector representation is rearranged and fed to bidirectional GRU to access forward and backward semantic features respectively. Finally, the latent topic attention is applied to strengthen the topic information in the bidirectional contextual features, and the bidirectional features are fused to generate the final text vector. Softmax classifier is used to give the sample label.

[Results] The accuracy of proposed model in three Chinese question datasets is 93.89%, 94.47% and 94.23% respectively, with the highest improvement of 5.82% over LSTM and of 4.5% over CNN.

[Limitations] We only use three Chinese question corpus to verify the performance of the model.

[Conclusions] This model can make full use of the semantic features of question texts, effectively overcoming the inaccurate understanding of the intention of questions, and improve the performance of question classification.

Key words Question classification      Multi-channel      Hierarchical attention      Convolution      GRU      
Published: 21 May 2020
ZTFLH:  TP391  

Cite this article:

YU Bengong, ZHU Mengdi. Question Classification Based on Hierarchical Attention Multi-Channel Convolution Bidirectional GRU . Data Analysis and Knowledge Discovery, 0, (): 1-.

URL:

http://manu44.magtech.com.cn/Jwk_infotech_wk3/EN/10.11925/infotech.2096-3467.2019.1292     OR     http://manu44.magtech.com.cn/Jwk_infotech_wk3/EN/Y0/V/I/1

[1] Xu Yuemei,Liu Yunwen,Cai Lianqiao. Predicitng Retweets of Government Microblogs with Deep-combined Features[J]. 数据分析与知识发现, 2020, 4(2/3): 18-28.
[2] Xiang Fei,Xie Yaotan. Recognition Model of Patient Reviews Based on Mixed Sampling and Transfer Learning[J]. 数据分析与知识发现, 2020, 4(2/3): 39-47.
[3] Bengong Yu,Yumeng Cao,Yangnan Chen,Ying Yang. Classification of Short Texts Based on nLD-SVM-RF Model[J]. 数据分析与知识发现, 2020, 4(1): 111-120.
[4] Weimin Nie,Yongzhou Chen,Jing Ma. A Text Vector Representation Model Merging Multi-Granularity Information[J]. 数据分析与知识发现, 2019, 3(9): 45-52.
[5] Yunfei Shao,Dongsu Liu. Classifying Short-texts with Class Feature Extension[J]. 数据分析与知识发现, 2019, 3(9): 60-67.
[6] Kan Liu,Lu Chen. Deep Neural Network Learning for Medical Triage[J]. 数据分析与知识发现, 2019, 3(6): 99-108.
[7] Yuemei Xu,Sining Lv,Lianqiao Cai,Xiaoya Zhang. Analyzing News Topic Evolution with Convolutional Neural Networks and Topic2Vec[J]. 数据分析与知识发现, 2018, 2(9): 31-41.
[8] Xiaoxi Huang,Hanyu Li,Rongbo Wang,Xiaohua Wang,Zhiqun Chen. Recognizing Metaphor with Convolution Neural Network and SVM[J]. 数据分析与知识发现, 2018, 2(10): 77-83.
  Copyright © 2016 Data Analysis and Knowledge Discovery   Tel/Fax:(010)82626611-6626,82624938   E-mail:jishu@mail.las.ac.cn